Content Hooks

content-hooks

generates distribution assets for every medium from a transcript and its research. tweets are written in full. everything else is hooks and titles. all items are scored with ICE.

inputs

• file: path to a transcript .md file, or the folder name inside notes-processing/
• research.md must already exist at notes-processing/{filename}/research.md
• output: notes-processing/{filename}/hooks.md

voice

all copy must sound like rami. read /Users/rami/Documents/life-os/ai-agents-config/skills/rami-voice/SKILL.md for the full rules. critical rules that apply to every hook and tweet you write:

• no em dashes — use a comma, colon, parentheses, or a new sentence instead
• no semicolons
• no hedging qualifiers: "it's worth noting", "one could argue", "it's interesting that"
• no corporate vocabulary: leverage (as verb), utilize, implement, unpack, delve, game-changing
• no summary conclusions — every closer should be a charge or a sharp stop, never a recap
• no formal transitions
• lowercase, no hashtags, no engagement bait ("drop it in the comments", "have you ever...")
• takes a side — observation is not enough, a take is required

preprocessing

clean the transcript before using it:

• strip YAML frontmatter, wiki-links ([[Note|text]] → text), and ## Related Notes section
• identify the dominant 1-2 themes (what the speaker kept coming back to) — use these as the hook lens

steps

1. read the cleaned transcript and research.md
2. identify the 1-2 dominant themes — these are the lens for all hooks
3. do 3-5 web searches to discover current trends:
   • what's trending right now in this space on LinkedIn / Twitter / YouTube
   • recent viral content on these exact themes
   • what's resonating with young founders and entrepreneurs this week
4. generate all content assets (see mediums below)
5. score each item with ICE
6. write hooks.md

mediums

linkedin — 5 hooks (opening lines only, not full posts)

• one line that stops the scroll
• counterintuitive, bold, or uncomfortably personal
• no "have you ever...", no rhetorical questions leading with "you"

twitter/x — 5 full tweets (written in full — these are final, not just hooks)

• naval ravikant / dan koe style: expose a truth that sounds simple but takes paragraphs to fully explain
• lowercase, no hashtags, no emojis, no engagement bait
• hard limit: 1-3 sentences only. if it's longer, cut it. tweets are not essays — they're the sharp point before the essay.
• each tweet must stand alone as a complete thought. no thread prefixes, no "1/n"
• examples that hit the right length:
  • "desire is a contract you make with yourself to be unhappy until you get what you want." — 1 sentence, reframes a universal experience
  • "you don't have a focus problem. you have a values problem." — 2 sentences, forces self-reflection
  • "i can indulge now, i'll be disciplined later — that's not a productivity strategy. that's addiction logic with better branding." — 2 sentences, uncomfortable and undeniable
  • "researchers found people choose a $20 task over a $25 one just because it has a shorter deadline. we do the same with our lives. we call it productivity." — 3 sentences, max length
• rami's voice: personal, direct, lowercase, ends sharp

instagram/tiktok — 5 hooks + 5 headline options

• hook: the first spoken sentence of a short video (grabs in 2 seconds)
• headline: text overlay or caption title (5-8 words, punchy)

youtube — 3 title + subtitle pairs + 1-2 sentence description each (2 for short-tier transcripts)

• title: searchable and compelling (not clickbait)
• subtitle: expands on the title, adds intrigue or specificity
• use section format in output (not a table — descriptions are too long for table cells)

ice scoring

score each item individually. be honest — score based on evidence from trend research, not optimism.

• impact (1-10): potential reach and resonance with rami's audience (young founders, entrepreneurs, gen z)
• confidence (1-10): how likely this lands, based on current trends and rami's track record with similar content
• ease (1-10): how easily produced from existing transcript material (10 = fully covered)
• ice = (impact + confidence + ease) / 3

medium ice = average of all items in that medium overall content ice = average of all medium ices

output format

# hooks: {transcript title}

**overall ice score:** {x.x}/10

---

## linkedin hooks

| # | hook | I | C | E | ICE |
|---|------|---|---|---|-----|
| 1 | hook text | 8 | 7 | 9 | 8.0 |

**medium ice:** {x.x}/10

---

## twitter/x

**1.** tweet text here
`I: 9 | C: 8 | E: 9 | ICE: 8.7`

**2.** tweet text here
`I: 8 | C: 9 | E: 9 | ICE: 8.7`

*(repeat for all 5)*

**medium ice:** {x.x}/10

---

## instagram/tiktok

### hooks
| # | hook | I | C | E | ICE |
|---|------|---|---|---|-----|

### headlines
| # | headline | I | C | E | ICE |
|---|----------|---|---|---|-----|

**medium ice:** {x.x}/10

---

## youtube

### 1. {title}
**subtitle:** {subtitle}
**description:** {1-2 sentences}
`I: 9 | C: 9 | E: 9 | ICE: 9.0`

*(repeat for all 3)*

**medium ice:** {x.x}/10

---

## trend context

- {bullet: what's trending that informed these hooks, with source url}

note on short transcripts

if the transcript is thin (< 150 words with 1-2 ideas), generate fewer hooks per medium (3 instead of 5) and focus them tightly on the one core insight. don't pad.